Data: Postcode Traits

In the endless search for tighter targeting, behavioural analysis has entered the data modelling game, but does it work?

What's their income and what's their postcode? Despite all the sophistication on hand for marketers, for the last 30 years virtually all direct mail has been targeted answering these two questions. Lifestyle and postcodes enjoy a theoretical buy-in that is rarely questioned. Individuals tend to cluster in the geographical areas they can afford - and, say marketers, direct mail results speak for themselves.

But data modelling is now attempting to break into virgin ground - the third and some say less predictive world of behavioural analysis. Few would be raising an eyebrow though if it wasn't for the fact that behaviour is also being couched as highly related to traditional postcode/income analysis. If postcode and income were marketers' 2D outlook of the world, postcode-related behaviour is in 3D.

The relationship is being made by Acxiom, which last month launched Personicx, its customer segmentation tool that groups households or postcodes into between 50-60 similar clusters that share behavioural characteristics. And intensity in this space is set to increase, with Experian next month launching Prime Performance Modelling (see next page), although its model is very different. Rather than use behaviour as a tool to find the prospect as Acxiom is, it is using its behavioural data sources to append selections already made. So can behaviour be used as a element equal to or more important than postcode, and does the link between them exist?

Anecdotally at least, there are reasons why income and postcode groupings aren't accurate. High earning plumbers, for example, can now be found living in postcodes shared by lawyers. Although their income score might say one thing, they aren't in the market for the same goods.

"The breakdown of classifications is fuelling a desire to come up with better segmentation ," says Simon James, head of data strategy at Zalpha.

According to James, whenever Zalpha has tested values, it has never been as conclusive as DM truisms - like lists of the most mail-responsive people.

"This is what marketers really want," he argues. "Our view is that values and behaviours don't tell you who to mail. They might indicate what to say on different versions of the mailpack, but that's all."

Postcode and behaviour link

Acxiom sees things very differently. Personicx product development manager James Horsburgh says he was as much surprised by the link between postcode and behaviour as anyone else, but says the link is robust. By 'behaviour' Horsburgh says he really means 'values' - whether they fly long-haul, how quick people adapt to technology, whether people prefer to pay with credit or cash. "When Acxiom bought Claritas and Consodata, we had some 350 of these datasets. We wanted to see if targeting this way created an new action, to push it and see if it worked."

According to Horsburgh, behaviour influenced the creation of clusters first. "Take one group we call 'spenders and borrowers'," he says. "They spend a lot and may look like they have high disposable income but they could have different demographics - they could be households with children or single occupiers, but the common fact is that they use credit. Behaviour comes first; we haven't forced the demographic element." That said, by championing behaviour first, he argues a link between it and postcode does appear.

Increased performance

"It's much more predictive than we first thought," he claims, citing a 28 per cent increase in performance (response of four to eight per cent) when compared to PRIZM. "There is a strong geo-influence on the way we lead our lives, especially in the countryside."

The link - and indeed the product - is not without its critics. Many argue it's simply PRIZM repackaged, with no new data sources, other than the lifestyle universe data which people can already use. "The issue with all modelled data is how far you can stretch an already stretched model," argues David Lovie, managing director at Data Vantage. "Marketing people all seem to be keen on behavioural data, but data specialists are not. They prefer a complete customer view utilising existing data, not this wider tool that doesn't offer much depth."

At a core level is a much fiercer debate about who defines a value, a lifestyle and a behaviour. "Lifestyle data already exists in great quantities, and we have our own with MOSAIC," explains Patrice Bendon, senior product manager at Experian. "We don't classify lifestyle as behavioural data. Getting a universal view of people's values is not easy." Acxiom does use its lifestyle surveys as the basis for proving behavioural and therefore values data. Is this an assumption too far?

"Personicx is designed to replace PRIZM and they do play a similar role," admits Horsburgh. "For example, having internet access is not automatic , and it reflects a person's attitude to technology. That's how we see this as a values-based measurement."

This certainly gets around the problem of creating a universe of values - something which previous suppliers have been limited with. The Values Company, for example, while arguably much deeper and more bespoke, is only filled out by a panel of 10,000 people and is difficult to roll out nationally. However some, Suzanne Lewis from HLB included, argue PRIZM is only suited to working with house files rather than cold lists. "If it's used for targeting with cold lists, I only see it going against MOSAIC, and question if it is really anything new," she says.

It's whoever shares this stark view that will decide if this new approach to segmentation is more spin than gain. "The Holy Grail of all marketing is the 'who', 'what' and 'why' elements," argues Stuart Broughton, director of Datalytics and formerly director at Claritas. "Who is answered by lifestyle, what is answered by transactional data and the why is the behavioural part. Attitudinal data will also answer the why, but you have to ask if this can really be implied. They've not interviewed the whole of the UK population. We answer the 'why' much better when a client has done their own market research and we overlay this with the lifestyle universe."

And just because you may be able target based on attitudes will clients be that interested if it only throws up the same people that would have been found through traditional segmentation? "You've got to use it differently if you use it at all," argues Suzanne Soper, executive director at Wegener, "but the bigger question is whether you know what you are getting with modelled data." She adds: "What marketers need to demand of segmentation is how much is modelled and how much is real. We have our own REALWealth and REALFile products with 42 million names, although only 13 million are real - and we're upfront about that."

But behavioural segmentation could be developing new fans. "In my experience, the defining factors are age and income, but this is definitely something we'd look to test," says Nick Davies, senior data planner at Broadsystem.

And others agree new systems can only improve data selection. "We can no longer use broad demographics," argues David Cole, director at CCB. "The chance of people living the same lifestyle is just too remote. Data collection needs to be more up-to-date."

Internet search engine Lycos is one unlikely provider that is making a foray into this area. It recently launched a little talked about service that collects real time databases through pop-up questions and sells these lists. It has already collected more than 20,000 attitude-based leads.

What's clear is that behavioural segmentation is a concept that will need a lot more than protestations about how successful it might be to persuade marketers to change the habits of a lifetime. But for Lovie, marketers should first heed some more simple advice: "People should put their own house in order, and segment their own data before using more cerebral tools."


Personicx: focuses on shared characteristics

Name: Personicx

What it is: Acxiom's new classification tool splitting households and postcodes according to their demographic, lifestyle and behavioural profiles.

Data Sources: Census 2001, consumer survey and lifestyle data from Claritas and Consodata and client customer files.

Options Available:

Personicx - 52 consumer classifications at household level that share similar demographic, lifestyle and behavioural characteristics. Tests show increase in performance of 28 per cent compared to household PRIZM.

GeoPersonicx - Segments the UK postcodes into 60 clusters - based on the majority behaviour of that postcode. Tests have shown 229 per cent performance increase over postcode level PRIZM.


Name: Prime Performance Modelling

What it is: Experian's data augmentation tool

Data Sources: Canvass Lifestyle, National Canvass, credit data and others.

Options Available:

First tier: Gives clients a profile of their file and finds lookalikes

Second tier: Segmentation of the customer file to find clusters that are modelled to find more niche consumers

Third Tier: Two-way segmentation with two models to find either repeat purchasers or not. All clusters found will vary client by client.


Before commenting please read our rules for commenting on articles.

If you see a comment you find offensive, you can flag it as inappropriate. In the top right-hand corner of an individual comment, you will see 'flag as inappropriate'. Clicking this prompts us to review the comment. For further information see our rules for commenting on articles.

comments powered by Disqus